C AHIERS DE
TOPOLOGIE ET GÉOMÉTRIE DIFFÉRENTIELLE CATÉGORIQUES
F. W ILLIAM L AWVERE More on graphic toposes
Cahiers de topologie et géométrie différentielle catégoriques, tome 32, no1 (1991), p. 5-10
<http://www.numdam.org/item?id=CTGDC_1991__32_1_5_0>
© Andrée C. Ehresmann et les auteurs, 1991, tous droits réservés.
L’accès aux archives de la revue « Cahiers de topologie et géométrie différentielle catégoriques » implique l’accord avec les conditions générales d’utilisation (http://www.numdam.org/conditions). Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
MORE ON GRAPHIC TOPOSES
by
F. William LAWVERECAHIERS DE TOPOLOGIE ET GÉOMÉTRIE DIFFÉRENTIELLE
CATÉGORIQUES
VOL. XXX II -1 (1991)
RÉSUMÉ. Un mondfde est
graphique
s’il satisfait A l’i- dentité deSchützenberger-Kimura
aba = ab . Dans cetarticle on continue l’itude des topos, dits topos gra-
phiques, engendr6s
par lesobjets
ayant un mondfde d’-endomorphismes
de ce type. On dimontre que dans untopos graphique,
n est uncogdn6rateur.
In
[Boulder AMS]
and[Iowa AMAST]
Ibegan
thestudy
of aspecial
class of combinatorial toposes. Thegeneral defining
property of these
special graphic
toposes is thatthey
aregenerated by
thoseobjects
whoseendomorphism
monoids arefinite and
satisfy
theSchützenberger-Kimura identity
aba =
ab ;
since this includes all localic(and
even all"one-way"
i.e. sites in which allendomorphism
monoids aretrivial) examples,
toemphasise
thespecial "graphic"
fea-tures, some results are stated for the case when a
single
such
object suffices,
inparticular,
when the topos ishyper-
connected. Monoids
satisfying
theSchutzenberger-Kimura identity
will also be called"graphic
monoids" forshort;
thefiniteness
assumption implies
thatthey
haveenough
con-stants. The other main consequence of the finiteness assump- tion is that the toposes under consideration are
actually presheaf toposes,
so that theprojectivity
of the generatorswill be used in calculations without further comment.
The
goals of
thisstudy
are two-fold: to illustrate in a"simple"
context sometopos-theoretic
constructions which arethus far difficult to calculate in
general,
and to arriveeventually,
viageometric realization,
at automaticdisplay
of
pictures
of hierarchical structures such as"hypercard", library catalogues,
etc.(The
actions ofnon-commuting
idem-potents are relevant to the latter
problem
in that x. a canbe
interpreted
as "the part of xreflecting a" ,
ratherthan
merely
"the part of x in a" as with commutativeintersection.
In this note I describe some results of thisstudy
which were obtained after theBangor meeting.
Among
the easypropositions
stated at themeeting
whichemphasize
the veryspecial
nature of agraphic
monoidM ,
are the facts that every left ideal is
actually
a two-sidedideal and that the lattice of all these is stable with res-
pect to the usual
right
action on all truth-values(= right ideals);
moreover, there is a latticeisomorphism
between the two-sided ideals and the essentialsubtoposes
of the topos of allapplications (=
d e fright actions)
ofM ,
with X-S c Xdefining
the "S-dimensional skeleton" of anyX ,
where S is agiven
left ideal of M . If theAufhebung
of S isdefined to be the smallest T;2 S such that every X. S is a
T-sheaf,
then theAufhebung
of 0 isT o -
the set of allconstants of
M ,
while theAufhebung
ofTo
isT 1 =
thesmallest left ideal which is connected as a
right
M-set(see [Iowa] Prop.1 ); X. T 1
has the same components as Xitself, justifying picturing
it as the "one-dimensional" ske- leton of X . Theputative picture
of ageneral
X is theinterlocking
union of thepictures
ofsingular figures
of thegenerating
form. Asignificant
three-dimensionalgenerating
form is the "taco" which
pictures
theeight-element graphic
monoid of all those endofunctors of
"any"
category 4 obtai- nedby considering
allcomposites
in any twiceaufgehobene adjoint analysis
of 4such as for
example,
an essentialanalysis
ofgraphic
toposeswith every S-skeleton a
T-sheaf,
and the empty M-set an S-sheaf.Since the set D of all
subtoposes
of a topos a is aco-Heyting algebra,
we can consider it as a closed category with T ~ Smeaning
T 2 S and ~meaning
u . For acombinatorial
topos
allsubtoposes
are essential and thusadmit a skeleton
functor,
so we can define a covariantfunctor
x display o cocomplete
D-enrichedcategories
by sending
X to the set X of allsubobjects
of X(using
direct
image
onmorphisms) with
the enrichmentand in
particular
X(O,B) = dim(B) .
Here we called the functor
"display
6tat0" ,
because al-though
itspecifies important
information about the desiredpicture,
the machine may still not know how to draw the lat- ter, unless wespecify
thepictures
of thegenerating forms;
the
examples
worked out in[Iowa AMAST]
show that these basicpictures
may betriangles,
but may also be squares or loz- enges, thatthey
may beclosed,
but may also bepartly
open,as for
example
the "taco" which has two two-dimensional closedsides,
but is open on top.One
example
where thedisplay
is well understood is the topos of reflexive directedgraphs,
which are theapplica-
tions of the three-element
graphic
monoid A 1 with two con-stants whose
generating
form is h->.
. Since this mon-oid may be considered to be the
theory
ofgraphs,
aprecise justification
for the term"graphic"
in the moregeneral
caseis
provided by
thefollowing
Proposition
1. Within the categoryof
allmonoids,
the smal- lestequational variety containing
thesingle
three-elementexample
e 1 is the categoryof
all monoidssatisfying
iden-tically
aba = ab .Proof It suffices to show that for each n , the free mon-
oid
Fn
in thesubcategory satisfying
theidentity
can beembedded in a
product
ofcopies
of e 1 ,Fn
->A1 In
asmonoids,
whereIn
is a set. The canonical choice forIn
is
In = (Fn,A1)
the set of monoidhomomorphisms
whichby
freeness is A1 as a set, for then there is a canonical hom-
omorphism Fn
->A1 A1 given by double-dualization,
which toeach
generating
letter ofFn assigns
thecorresponding
pro-jection n
-> Ai . It can be shown that this canonical hom-omorphism
isinjective,
i.e. that if u * w inFn ,
thenthere exists an x E
An1
such thatu(x) * w(x)
in A 1 . Here we use the fact that the elements u,w are words with-out
repetitions
in the nletters,
and thatw(x) =
thefirst of the two constants of A1 which occurs in the list
x o w .
Note that while maps of
applications in
a fixedgraphic
topos never increase dimension
(as
isexpressed
in the funct-orality
ofdisplay o above), by
contrast maps betweengraph-
ics often
do;
forexample,
the "taco" is ahomomorphic image
of
F6 ,
but any freegraphic
monoid is one- dimensional.Although
every topos is the fixedpart
of agraphic
top-os
(indeed
one of the formA 1 /X)
for some left-exact idem- potent functor on thelatter,
thegraphic
toposes are them- selvesspecial
in many respects. This is born outby
thefollowing proposition
whichgeneralizes
a remark madeby
Borceux many years ago. Recall that in any topos, n is an internal cogenerator in the
x sense
that anyobject injects
into some function space fix ;
however,
to get an actual cogenerator we must be able toreplace
the exponentby
a dis-crete one, and this
usually
does nothappen.
Proposition
2.(Gustavo Arenas)
In anygraphic
topos, n isa cogenerator.
Proof First note that
if A->X
can bedistinguished by
any map X ---7 n, then
they
can bedistinguished by (the
characteristic map
of)
someprincipal subobject (image
ofsome
singular figure
B->X) .
In thegraphic
case we can even take B =A ;
for if theendomorphisms
of Asatisfy
the
graphic identity
and x1,x2 are twofigures
in X ofform
A ,
consider the characteristic map[z]
of theimage
of z = x 1 or z = X2 - If
[z] x1
=[z]X2
for both thesez , then there exist
endomorphisms a,b
so that bothwhich
implies
The
computation
of theAufhebung
relation on essentialsubtoposes
has not yet been carried out in manyimportant examples
such asalgebraic
or differentialgeometry.
Even forgraphic
toposes thiscomputation
may be difficult. How- ever, in the latter case it can at least be reduced to aquestion involving
congruence relations on ideals of the mon-oid itself :
Proposition
3.If
S g T areleft
ideals in agraphic
mon-oid
M ,
then in the associatedgraphic
topos,
onehas the condition
"every
X. S is aT-sheaf’ if
andonly if for
every congruence relation = on T as aright
M-set onehas the
implication
Construing
the elements of the lattice D of essentialsubtoposes
as refineddimensions,
one would also like to add them. Since the infima in D do notalways
agree[Kelly- Lawvere]
withintersections,
it may be necessary ingeneral
to
replace
D with(2D)OP
to get a workabletheory.
Theidea is that the sum
S1*S2
of two dimensions should be the smallest T for whichholds for the skeleta of all
X,Y
in the topos.(It
is usu-ally only
for dimension0 ,
T =To ,
that() -
· T pre-serves
products). However,
forgraphics
the infima do agree with intersections and moreover we haveProposition
4.If S1,S2,T
areleft
ideals in agraphic monoid,
then the above relation holdsif
andonly if
Moreover, if T,T’
both have this property relative toS1,S2
then so does T nT’ ;
thus the intersectionof
allsuch T
gives
a well-defined
"sum"of S1*-S2 .
Proof The
necessity
of the condition follows from consider-ing X,Y
to beprincipal right
ideals.Conversely,
if itholds and
X E X.S1, y E Y · S2
for somearbitrary
M-applications X,Y
with Sl,S2witnessing
thatfact,
i.e.XS1 = X, yS2 = y , then
choosing
t as in the conditiongives
that x,y> is in(X
xY) · T
becauseIf u E T’
(for
some other suchT’)
also fixes agiven
pair
S1,S2 that t e Tfixes,
then tu E T n T’ since these areideals,
andobviously
showing
that T n T’ serves as well or better.One has
To*S
= S for allS ,
but clarification is needed on theQuestion.
What is the relation between the "successor"S. T 1
and the
Aufhebung
S’ of S forgraphic toposes?
Recallthat Michael Zaks
(unpublished)
showed for thesimplicial
topos that while these two constructions
give equal
resultsfor S =
To,T1,T2, by
contrast for S 2T 3
one has ins-tead that the doubled dimension S*S =
S’*Ti
for S’ the(smallest) Aufhebung
ofS ,
i.e.(since
dimensions reduceto natural numbers
together
with ± oo in thesimplicial case) that
n’ = 2n - 1 for n > 3 .REFERENCES
F.W.LAWVERE, Qualitative
distinctions between some toposes ofgeneralized graphs, Proceedings of
AMS Boulder 1987Symposium
onCategory Theory
andComputer Science, Contemporary
Mathematics 92(1989)
261-299.G.M. KELLY & F.W.
LAWVERE,
On the,Complete
Lattice of EssentialLocalizations,
Bull. SociétéMathematique
deBelgique,
XLI(1989)
289-319.F.W.
LAWVERE, Display
ofGraphics
and theirApplications,
asExemplified by 2-Categories
and theHegelian
"Taco",Proceedings of
the First InternationalConference
onAlgebraic Methodology
andSoftware Technology,
TheUniversity of Iowa, (1989)
51-75.(There
is an error in the version of the latter paper distributed at theBangor Meeting:
Freegraphic
monoids donot act
faithfully
on theirconstants.)
December
14, 1989,
revisedAugust
7, 1990F. William Lawvere SUNY at Buffalo 106 Diefendorf Hall
Buffalo,
NY 14214USA